Physicians, both directly through services they provide and indirectly through the services that they order, shape the vast majority of health care utilization in the US. There has long been intense interest in the determinants of physician decision making because optimizing physician decisions has the potential to improve the quality of US health care while also controlling costs by eliminating the use of unnecessary services. One influence on physician decision making that has largely been ignored derives from the influence of networks of physicians on individual physician decision making. Here, we are using the term network in a technical sense to refer to informal networks of primary care and specialist physicians who share patients and information, rather than formal networks of all of the physicians formally affiliated with a health plan or hospital. By virtue of sharing patients, these networks can be defined empirically. Networks of physicians have the potential to influence an individual physician's decision making, but such influences have rarely been studied. In this application, we propose to develop a longitudinal database of physician networks over time and to apply the methods we have developed in defining and describing physician networks to understand diffusion of cancer care practices across these networks. We propose to use Medicare data from 2005 through 2009 to evaluate the contribution of physician networks to variations in the adoption and use of innovations in oncology care as well as hospice use for terminally-ill cancer patients. We also plan additional analyses that will combine data from the Surveillance, Epidemiology and End Results (SEER) Program with our data on physician networks in those areas to examine the appropriateness of the adoption and use of new cancer treatments. The specific aims of the proposal are to: 1) Examine the extent to which network properties predict diffusion of new cancer therapies, focusing on the diffusion of new biologic therapies for cancer; 2) Examine the extent to which network properties predict local variations in the use of hospice; and 3) Examine the extent to which network properties predict the appropriateness of the diffusion of new therapies.